Multi-scale modeling and verification method and system based on ocean current spring layer structure analysis

Abstract

The invention relates to the technical field of ocean current data analysis, in particular to a multi-scale modeling and verification method and system based on ocean current spring layer structure analysis. And obtaining the thermocline and halocline parameters of the target sea area and carrying out abnormal value elimination and spatial interpolation to form a thermocline data set. A convolutional neural network is used to extract spring layer space structure features, and based on Gaussian curvature quantification boundary complexity, a thermocline and halocline equivalent geometric model is constructed. On the basis, taking the equivalent geometric model as a boundary condition, and performing multi-scale decomposition on the Navier-Stokes equation set to generate a dynamic approximate equation; and then inputting the approximate equation and the equivalent geometric model into a physical information neural network for constraint training to obtain a multi-scale flow field prediction model, and comparing the multi-scale flow field prediction model with actually measured flow field data to complete verification. According to the method, automation, refinement and multi-scale coupling of spring layer structure modeling are achieved, and scientificity and engineering applicability of ocean current modeling are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ocean current data analysis, and particularly relates to a multi-scale modeling and verification method and system based on ocean current thermocline structure analysis. BACKGROUND

[0002] In related technologies, as a key component of the marine power system, ocean currents play an important role in global climate regulation, material transport, and marine ecosystem evolution. The vertical structure of ocean currents is significantly affected by temperature and salinity distribution, especially in the thermocline region, where there is a clear temperature and salinity jump phenomenon, forming a complex multi-scale dynamic process. The thermocline is not only an important boundary of ocean mixing, but also a key interface for energy, momentum and mass exchange, playing a key role in regulating ocean current evolution and energy transfer.

[0003] However, the existing ocean current modeling method has the following shortcomings in dealing with the multi-scale structure characteristics of the temperature and salinity thermocline region: traditional thermocline structure extraction methods mostly rely on empirical models or rule thresholds, which are difficult to accurately depict the spatial morphological changes of complex thermoclines, resulting in unclear boundary identification; secondly, in the marine environment, there are complex processes such as temperature and salinity interaction, Gaussian curvature mutation, and energy interlayer transfer, and modeling methods driven by a single scale or a single physical quantity cannot fully reflect the real flow field distribution under the thermocline; in addition, existing flow field prediction models are mostly based on black-box neural networks, ignoring the physical constraints of fluid mechanics equations, resulting in weak physical interpretability and poor generalization ability of the prediction results.

[0004] In summary, the technical problems in related technologies need to be improved. SUMMARY

[0005] The main purpose of the embodiments of the present application is to propose a multi-scale modeling and verification method and system based on ocean current thermocline structure analysis, to realize the automation, refinement and multi-scale coupling of thermocline structure modeling, and to improve the scientificity and engineering applicability of ocean current modeling.

[0006] To achieve the above-mentioned purpose, one aspect of the embodiments of the present application proposes a multi-scale modeling and verification method based on ocean current thermocline structure analysis, which comprises: obtaining parameter data of the temperature and salinity thermocline region of a target sea area; performing outlier rejection and spatial interpolation on the parameter data to form a temperature and salinity thermocline dataset; based on the temperature and salinity thermocline dataset, using a convolutional neural network to extract temperature and salinity thermocline spatial structure features; According to the temperature and salinity thermocline spatial structure features, the complexity of the temperature and salinity thermocline boundary is quantified by the Gaussian curvature parameter, and an equivalent geometric model of the temperature and salinity thermocline is established; With the equivalent geometric model of the thermocline as a boundary condition, the Navier-Stokes equation set is multi-scale decomposed to generate a multi-scale dynamic approximation equation; The multi-scale dynamic equation is input into a physical information neural network with the equivalent geometric model of the thermocline as a constraint to train and generate a multi-scale flow field prediction model; Based on the multi-scale flow field prediction model, the flow velocity and direction of a target sea area are simulated and calculated, and compared with the measured flow field data to complete multi-scale modeling and verification.

[0007] In some embodiments, the parameter data includes temperature distribution, salinity gradient, sea current velocity and depth information.

[0008] In some embodiments, based on the thermocline data set, a convolutional neural network is used to extract the spatial structure features of the thermocline, including: The thermocline data set is grid pretreated, and the temperature field and salinity field are constructed into a multi-channel input tensor according to the depth level; A convolutional neural network is used to perform multi-scale convolution on the multi-channel input tensor to obtain local spatial features representing the gradient change of the thermocline; According to the local spatial features, a feature mapping sequence is generated based on the convolution output after a pooling operation and a feature compression operation; The feature mapping sequence is input into a fully connected structure to generate spatial structure features.

[0009] In some embodiments, the spatial structure features include thermocline boundary position, gradient intensity distribution and thermocline thickness change information.

[0010] In some embodiments, based on the thermocline spatial structure features, the complexity of the thermocline boundary is quantified by a Gaussian curvature parameter, and an equivalent geometric model of the thermocline is established, including: The spatial structure features are discretely reconstructed to obtain an initial surface of the thermocline boundary; Based on the initial surface, the Gaussian curvature of each discrete point is calculated, the curvature abnormal points are smoothed and corrected to generate a thermocline boundary curvature distribution; The complexity of the thermocline boundary is quantified according to the thermocline boundary curvature distribution; The quantification result is used as an equivalent geometric constraint to construct an equivalent geometric model of the thermocline; the equivalent geometric model is used to represent the equivalent bending degree of the thermocline boundary and its spatial variation trend.

[0011] In some embodiments, the formula of the Gaussian curvature is as follows: ; Where G represents the Gaussian curvature of the thermocline interface. represents the first order spatial variation rate of the thermocline interface in the longitude direction; represents the first order spatial variation rate of the thermocline interface in the latitude direction; represents the second order curvature of the thermocline interface along the longitude direction; represents the second order curvature of the thermocline interface along the latitude direction; represents the mixed curvature term of the thermocline interface in the longitude-latitude direction.

[0012] In some embodiments, the multi-scale decomposition of the Navier-Stokes equations with the thermocline equivalent geometric model as the boundary condition generates a multi-scale dynamic approximation equation, including: Based on the curvature distribution of the thermocline equivalent geometric model, the Navier-Stokes equations are divided into a local flow scale and a regional overall flow scale in the thermocline neighborhood; In the local flow scale, the interface curvature feature, the thermocline gradient variation, and the thermocline thickness parameter are introduced as boundary disturbance source terms into the mechanical equation to construct a multi-scale approximation structure; the multi-scale approximation structure includes a zero-order background flow, a first-order correction flow dominated by interface curvature, and a high-order correction flow driven by thermocline gradient; In the regional overall flow scale, based on the multi-scale approximation structure, the multi-scale dynamic approximation expression is obtained with the Coriolis force parameter, the scale ratio parameter, and the gradient feature as the dominant small parameters; the multi-scale dynamic approximation expression includes the effects of the earth's rotation, the boundary geometry, and the thermocline stratification; Based on the multi-scale dynamic approximation expression, a multi-scale dynamic approximation equation is formed; the multi-scale dynamic approximation equation includes a Coriolis force term and a thermocline coupling term.

[0013] In some embodiments, the formula of the multi-scale approximation structure is as follows:

[0014] wherein, represents the total velocity field of the ocean flow field; represents the zero-order dominant flow field; represents the first-order correction flow field; represents the second-order correction flow field; represents the Coriolis force parameter.

[0015] In some embodiments, the multi-scale dynamic equation and the thermocline equivalent geometric model are input into a physical information neural network for constrained training to generate a multi-scale flow field prediction model, including: The term structure of the multi-scale dynamic equation is discretized and encoded in the form of physical residual to input the physical information neural network; Based on the spatial structure characteristics of the thermocline equivalent geometric model, the temperature field and salinity field observation data are normalized in the same scale as the data constraint input of the physical information neural network; By iteratively optimizing the weight parameters of the physical information neural network, a multi-scale flow field prediction model is obtained.

[0016] To achieve the above object, another aspect of the embodiment of the present application proposes a multi-scale modeling and verification system based on ocean current thermocline structure analysis, which comprises: A data acquisition module is configured to acquire parameter data of a target sea area thermocline region; A preprocessing module is configured to perform outlier rejection and spatial interpolation on the parameter data to form a thermocline data set; A feature extraction module is configured to extract thermocline spatial structure features based on the thermocline data set using a convolutional neural network; A first construction module is configured to quantify the complexity of the thermocline boundary by a Gaussian curvature parameter according to the thermocline spatial structure features, and establish a thermocline equivalent geometric model; A multi-scale decomposition module is configured to perform multi-scale decomposition on the Navier-Stokes equation set with the thermocline equivalent geometric model as the boundary condition, and generate a multi-scale dynamic approximation equation; A second construction module is configured to input the multi-scale dynamic equation and the thermocline equivalent geometric model into a physical information neural network for constraint training, and generate a multi-scale flow field prediction model; A verification module is configured to simulate and calculate the flow velocity and direction of the target sea area based on the multi-scale flow field prediction model, and compare with the measured flow field data to complete multi-scale modeling and verification.

[0017] To achieve the above object, another aspect of the embodiment of the present application proposes an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0018] To achieve the above object, another aspect of the embodiment of the present application proposes a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above method.

[0019] To achieve the above object, another aspect of the embodiment of the present application proposes a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the above method.

[0020] The embodiment of the present application at least has the following beneficial effects: the present application provides a multi-scale modeling and verification method and system based on ocean current thermocline structure analysis, which obtains parameter data of the target sea area thermocline region, and performs outlier rejection and spatial interpolation, effectively improving the data quality and providing a reliable basis for subsequent modeling; secondly, the spatial structure features of the thermocline are automatically extracted by using the convolutional neural network, avoiding the limitations of artificial feature design, and enhancing the expression ability of complex thermocline morphology; further, the Gaussian curvature parameter is used to quantify the complexity of the thermocline boundary, and an equivalent geometric model is constructed to provide accurate boundary conditions for multi-scale dynamic modeling; in addition, the Navier-Stokes equation set is decomposed in multiple scales, and the input physical information neural network is trained in combination with the thermocline structure constraint to realize a multi-scale flow field prediction model with strong physical consistency and high calculation efficiency; finally, the prediction result is compared and verified with the measured data to ensure the accuracy and practicability of the model. The present application realizes the automation, refinement and multi-scale coupling of thermocline structure modeling, and improves the scientificity and engineering applicability of ocean current modeling. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of a multi-scale modeling and verification method based on ocean current thermocline structure analysis provided by the embodiment of the present application; Figure 2 is a flowchart of step S5 in Figure 1 ; Figure 3 is a comparison chart of u, v prediction data and real data of the physical information neural network prediction provided by the embodiment of the present application; Figure 4 is a module schematic diagram of a multi-scale modeling and verification system based on ocean current thermocline structure analysis provided by the embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application, but are only examples of devices and methods consistent with some aspects of the embodiments of the present application as described in detail in the appended claims.

[0023] It can be understood that the terms "first", "second", and the like used in the present application can be used herein to describe various concepts, but unless specifically stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiments of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon determining" or "in response to determining".

[0024] The terms "at least one", "multiple", "each", "any" and the like used in the present application include one, two or more than two, multiple includes two or more than two, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0026] Before the embodiments of the present application are described in detail, first, some nouns and terms involved in the embodiments of the present application are described, and the nouns and terms involved in the embodiments of the present application are applicable to the following explanations.

[0027] Thermocline: an important water layer structure in the ocean, referring to the region where the temperature and salinity of seawater change sharply with the depth of the ocean.

[0028] Navier-Stokes equations: abbreviated as NS equations, a partial differential equation set used to describe the motion law of viscous incompressible fluid, which can represent the evolution law of physical quantities such as fluid velocity and pressure with time and space.

[0029] Coriolis force: refers to the inertial force generated by the rotation of the earth, which is one of the key factors affecting the motion of geophysical fluids (such as ocean, atmosphere).

[0030] Gaussian curvature: a geometric parameter describing the bending degree of a surface at a point, used to quantify the "equivalent bending degree" of the thermocline boundary (the greater the temperature and salinity gradient, the greater the equivalent Gaussian curvature).

[0031] The embodiment of the application provides a multi-scale modeling and verification method and system based on ocean current jump layer structure analysis, which obtains parameter data of a temperature-salinity jump layer region of a target sea area, performs outlier rejection and spatial interpolation, effectively improves data quality, and provides a reliable basis for subsequent modeling; secondly, the spatial structure features of the temperature-salinity jump layer are automatically extracted by using a convolutional neural network, avoiding the limitations of artificial feature design and enhancing the expression ability of complex jump layer morphology; further, the jump layer boundary complexity is quantified by using a Gaussian curvature parameter, and an equivalent geometric model is constructed, thereby providing accurate boundary conditions for multi-scale dynamic modeling; in addition, the Navier-Stokes equation set is decomposed in multiple scales, and a physical information neural network is trained in combination with jump layer structure constraints, thereby realizing a multi-scale flow field prediction model with strong physical consistency and high calculation efficiency; finally, the prediction result is compared and verified with measured data, thereby ensuring the accuracy and practicability of the model. The scheme realizes automation, refinement and multi-scale coupling of jump layer structure modeling, and improves the scientificity and engineering applicability of ocean current modeling.

[0032] The multi-scale modeling and verification method based on ocean current jump layer structure analysis provided by the embodiment of the application relates to the technical field of ocean current data analysis. The multi-scale modeling and verification method based on ocean current jump layer structure analysis provided by the embodiment of the application can be applied to a terminal, can also be applied to a server, and can be software running in the terminal or the server. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a vehicle-mounted terminal, and the like, but is not limited thereto; the server end can be configured as a stand-alone physical server, can also be configured as a server cluster or a distributed system composed of multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform, and the server can also be a node server in a blockchain network; the software can be an application of the multi-scale modeling and verification method based on ocean current jump layer structure analysis, and the like, but is not limited to the above forms.

[0033] The application is operable with numerous general purpose or special purpose computer system environments or configurations. Examples include: personal computers, server computers, hand held or laptop devices, tablet devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The application can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, that perform particular tasks or implement particular abstract data types. The application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in local and remote computer storage media including memory storage devices.

[0034] Referring to Figures 1-4 The application relates to a multi-scale modeling and verification method and system based on ocean thermocline structure analysis.

[0035] Figure 1 is an optional flowchart of a multi-scale modeling and verification method based on ocean thermocline structure analysis provided by the application, Figure 1 The method in the application can include but is not limited to steps S1 to S7: S1: obtaining parameter data of a target sea area thermocline region; the parameter data includes temperature distribution, salinity gradient, current velocity and depth information.

[0036] In the embodiment, the eastern Mediterranean Sea is selected as a target research area, the longitude range of which is 23°E-28°E, the latitude range of which is 34°N-36°N, the depth range of which is 40 m-650 m, and the time span of which is from March 1, 2023 to September 30, 2023. The temperature thermocline and salinity thermocline structure of the area are obvious, the ocean current intensity is representative, and the area is suitable as a multi-scale modeling test area.

[0037] Four types of original parameter data, including temperature profile, salinity profile, three-dimensional current velocity distribution and corresponding depth coordinate information, are downloaded through a marine reanalysis data platform. The data contains longitude and latitude grids, vertical depth grids and time series of each day or each hour.

[0038] After downloading, the data is standardized according to the spatial dimensions (longitude, latitude and depth), and clustering analysis is performed on multiple profiles at the same time to reduce local abnormal deviations caused by observation noise, so as to obtain a smooth basic data set that can be used for subsequent feature analysis.

[0039] S2: Outlier removal and spatial interpolation are performed on the parameter data to form a thermohaline jump dataset; In this embodiment, quality control is performed on the acquired parameter data.

[0040] First, a density distribution-based anomaly detection method is used to identify measurement errors, such as temperature spikes or salinity drops in certain depth layers that are significantly inconsistent with the surrounding continuous layers. The detected outliers are replaced with multi-dimensional neighborhood average values.

[0041] Subsequently, to obtain continuous jump morphology, three-dimensional spatial interpolation is performed on the cleaned data. During the interpolation process, horizontal interpolation is performed along the longitude and latitude directions, respectively, and vertical interpolation is performed along the depth direction, so that the temperature and salinity gradients remain continuous in space, and finally a regularized three-dimensional thermohaline jump dataset is formed.

[0042] This dataset contains complete temperature variation trends, salinity gradient characteristics, and jump transition zones with depth, and can be directly used as input for convolutional neural networks.

[0043] S3: Based on the thermohaline jump dataset, a convolutional neural network is used to extract thermohaline jump spatial structure features; Wherein, based on the thermohaline jump dataset, a convolutional neural network is used to extract thermohaline jump spatial structure features, comprising: S31: Grid preprocessing is performed on the thermohaline jump dataset, and the temperature field and salinity field are constructed into a multi-channel input tensor according to the depth level; In this embodiment, the thermohaline jump dataset is grid preprocessed, and the longitude, latitude and depth directions are divided into regular grids, so that the ocean profile is discretized into three-dimensional voxel blocks.

[0044] Subsequently, according to the depth layering, the temperature value and salinity value of each layer are combined into two independent channels to form a multi-channel input tensor. The structure of the input tensor is: Height dimension: corresponding to the latitude direction; Width dimension: corresponding to the longitude direction; Channel dimension: temperature field channel, salinity field channel; Depth dimension: stacked according to different depth layers of 50 m, 100 m, 150 m, etc.

[0045] In this way, the input tensor can completely retain the spatial continuity of the temperature variation trend, salinity gradient interval and jump region, laying a foundation for convolutional neural network cross-depth structure feature extraction.

[0046] S32: A convolutional neural network is used to perform multi-scale convolution on the multi-channel input tensor to obtain local spatial features representing the gradient variation of the thermohaline jump. In this embodiment, the front end of the convolutional neural network is designed as a multi-scale convolutional module composed of different size convolutional kernels, including fixed large convolutional kernels for capturing deep global trends and small convolutional kernels for capturing high gradient mutation regions.

[0047] The specific implementation includes: using small-scale convolutional kernels to identify the small differences in temperature / salinity values between adjacent grid points; using medium-size convolutional kernels to detect the rapidly changing bands in the thermocline; and using large-size convolutional kernels to extract the overall thermocline trend in a larger spatial range.

[0048] This multi-scale convolutional structure can detect in the local space: the location of rapid gradient change; the overlapping area of the temperature thermocline and the salinity thermocline; the continuous change of the thermocline thickness with latitude and longitude; and the boundary uplift and subsidence phenomenon caused by local mutations.

[0049] The convolutional layer outputs a set of primary local feature maps representing the change trend and complexity of each spatial position in the temperature and salinity channels.

[0050] S33: Based on the local spatial features, a feature mapping sequence is generated based on the convolution output after pooling operation and feature compression operation; S34: The feature mapping sequence is input into a fully connected structure to generate a spatial structure feature; the spatial structure feature includes the temperature-salinity thermocline boundary position, gradient intensity distribution, and thermocline thickness change information.

[0051] S4: According to the temperature-salinity thermocline spatial structure feature, the temperature-salinity thermocline boundary complexity is quantified by a Gaussian curvature parameter, and an equivalent geometric model of the temperature-salinity thermocline is established; The temperature-salinity thermocline spatial structure feature is quantified by a Gaussian curvature parameter, and an equivalent geometric model of the temperature-salinity thermocline is established, including: S41: Discrete surface reconstruction is performed on the spatial structure feature to obtain an initial surface of the thermocline boundary; In this embodiment, the position point set of the upper and lower boundaries of the thermocline is extracted from the spatial structure feature output by the convolutional neural network. The point set is represented in the form of "longitude-latitude-depth".

[0052] To reconstruct the discrete position points into a continuous surface: a uniform grid is constructed on the longitude-latitude plane, with each grid point corresponding to a predicted thermocline depth value; for points with missing or low confidence, the depth is filled in by neighborhood interpolation; and a surface fitting operation is performed on the overall point set to generate an initial surface of the thermocline boundary.

[0053] In this embodiment, the curved surface reconstruction result is a three-dimensional continuous curved surface, which is used to represent the spatial distribution of the pycnocline along the entire sea area, and can reflect the uplift area, sinking area and local mutation zone of the pycnocline. The curved surface serves as a reference geometry object for subsequent curvature analysis.

[0054] S42: Based on the initial curved surface, the Gaussian curvature of each discrete point is calculated, the curvature anomaly point is corrected by smoothing, and the pycnocline boundary curvature distribution is generated; In this embodiment, discrete curvature analysis is performed on the reconstructed initial curved surface, mainly following the following process: On the curved surface grid, the neighborhood points around each grid point are obtained, and the bending degree of the region is analyzed.

[0055] The calculation process focuses on the following types of local features: bending trend of the curved surface in the longitude direction; bending trend of the curved surface in the latitude direction; change of local uplift or sinking speed; contribution of gradient mutation section to curvature.

[0056] Finally, a “local Gaussian curvature value” is generated for each grid point, which is used to describe the equivalent bending strength of the point.

[0057] Due to the existence of noise and mutation characteristics in real ocean data and model prediction, in order to ensure the continuity of the curvature distribution, it is necessary to smooth the abnormal curvature points: for points that obviously deviate from the average value of the neighborhood, the neighborhood average method or local weighted smoothing method is used for correction; the necessary curvature peak value is retained in the high gradient mutation area to not destroy the physical meaning; the curvature information is filled in the data sparse area by depth direction interpolation method. After correction, a continuous and stable “pycnocline boundary curvature distribution map” can be obtained.

[0058] S43: According to the pycnocline boundary curvature distribution, the complexity of the pycnocline boundary is quantified; In this embodiment, the Gaussian curvature distribution is used to quantify the complexity of the pycnocline boundary, mainly embodied in: A larger curvature value means that the pycnocline boundary is strongly fluctuating and the local geometric structure is complex; a smaller curvature value means that the boundary is flat and changes gently.

[0059] If the curvature value changes rapidly in some areas, it indicates that there are geometric mutations in the pycnocline boundary at these positions; these areas often correspond to strong gradient changes or sea current deflection zones, which are important inputs for multi-scale dynamics.

[0060] Through statistical analysis of the curvature distribution, the following quantitative results can be obtained: the overall average complexity of the boundary; the spatial distribution of the most obvious fluctuation area; the trend of complexity change with latitude and longitude; the complexity gradient of depth. The quantitative results will be used as the core parameter input for constructing the equivalent geometric model.

[0061] S44: constructing an equivalent geometric model of the thermocline based on the quantification result as an equivalent geometric constraint; the equivalent geometric model is used to represent the equivalent bending degree of the thermocline boundary and the spatial variation trend thereof.

[0062] In the embodiment, an equivalent geometric model of the thermocline is constructed according to the curvature distribution and the complexity quantification result, and the specific implementation manner is as follows: The equivalent geometric model uses the curvature distribution to constrain the boundary shape, including: defining the upper limit and the lower limit of the boundary bending intensity to ensure the physical reasonableness of the model; enhancing the geometric stability of the model in the region where the boundary is relatively smooth; and increasing the sensitivity of the model to local structures in the region where the curvature is relatively large, so as to more accurately reflect the real characteristics of the thermocline shape.

[0063] According to the quantified complexity, the spatial distribution of the thermocline is divided into different levels of geometric structures: a low-complexity region adopts a smooth transition surface; a medium-complexity region adopts a layered transition structure; and a high-complexity region constructs an equivalent geometric shape with obvious bending characteristics.

[0064] The finally obtained equivalent geometric model can completely explain: the overall distribution trend of the thermocline boundary; the spatial bending degree caused by the curvature variation; the boundary lifting, sinking and local undulation; and the corresponding spatial continuity and stability.

[0065] Specifically, the formula of the Gaussian curvature is as follows: ; Wherein, G represents the Gaussian curvature of the thermocline interface; represents the first-order spatial variation rate of the thermocline interface in the longitude direction; represents the first-order spatial variation rate of the thermocline interface in the latitude direction; represents the second-order curvature of the thermocline interface along the longitude direction; represents the second-order curvature of the thermocline interface along the latitude direction; represents the mixed curvature term of the thermocline interface in the longitude-latitude direction.

[0066] S5: performing multi-scale decomposition on the Navier-Stokes equation set with the thermocline equivalent geometric model as a boundary condition to generate a multi-scale dynamics approximate equation; As shown in the reference Figure 2 , the multi-scale decomposition on the Navier-Stokes equation set with the thermocline equivalent geometric model as a boundary condition to generate a multi-scale dynamics approximate equation includes: S51: based on the curvature distribution of the thermocline equivalent geometric model, dividing the Navier-Stokes equation set into a local flow scale and a regional overall flow scale in the neighborhood of the thermocline. In this embodiment, the boundary curvature information, the pycnocline thickness information and their spatial distribution in the longitude and latitude directions corresponding to each grid point in the equivalent geometric model of the thermocline are read. Taking the pycnocline boundary as the center, a certain thickness of the neighborhood (for example, a range of several times the pycnocline thickness) above and below the pycnocline boundary is selected as a local flow scale, which is specially used to depict the detailed flow structure near the pycnocline caused by geometric bending, temperature-salinity gradient and density stratification; And the entire target sea area (for example, a specified sub-region of the Mediterranean Sea) is regarded as a regional overall flow scale, which is used to depict the overall flow direction and overall flow velocity variation under the action of large-scale background circulation structure, topography and geostrophic balance and other factors.

[0067] In the division process, the system automatically identifies regions with high boundary complexity according to the curvature distribution, and sets the regions near these regions as key calculation regions of the local scale. For regions with small curvature and gentle boundary, the background flow pattern is mainly used in the overall scale, and only necessary corrections are retained, forming a multi-scale solving framework in which the local scale and the overall scale are nested with each other.

[0068] S52: In the local flow scale, the interface curvature characteristics, temperature-salinity gradient variation and pycnocline thickness parameters are introduced as boundary disturbance source terms into the mechanical equation to construct a multi-scale approximation structure; the multi-scale approximation structure includes a zero-order background flow, a first-order correction flow dominated by interface curvature and a high-order correction flow driven by temperature-salinity gradient; In this embodiment, in the local flow scale, the boundary shape, curvature distribution, pycnocline thickness and temperature-salinity gradient information provided by the equivalent geometric model of the thermocline are taken as inputs, and these information are regarded as boundary disturbance source terms acting on the Navier-Stokes equation to construct a layered multi-scale approximation structure.

[0069] 1. Construction of zero-order background flow: In this step, it is assumed that there is no boundary fluctuation, temperature-salinity gradient and density stratification and other complex effects, and based on the large-scale background pressure field and the average wind stress condition, a relatively smooth zero-order background flow in the target sea area is obtained. The background flow describes the main flow direction and flow velocity level of the ocean current on a large scale when the details of the pycnocline are ignored, and is the basis part of the entire multi-scale approximation structure.

[0070] 2. First-order correction flow dominated by interface curvature: The curvature information in the equivalent geometric model of the thermocline is introduced into the local flow scale, and the bending degree of the boundary is regarded as a disturbance source of fluid motion. In the region near the pycnocline with large curvature, the first-order correction flow mainly embodies: the flow line is deflected and attached near the curved interface; a local acceleration or deceleration band along the interface direction appears; there is a slight shear difference between the upper and lower interfaces. By superimposing this first-order correction flow dominated by interface curvature on the basis of the zero-order background flow, a local flow field structure closer to the actual pycnocline geometric shape can be obtained.

[0071] 3. High-order correction flow driven by temperature and salinity gradients: Based on the first-order curvature correction, the spatial variations of temperature gradient, salinity gradient and pycnocline thickness are further introduced into the local dynamics structure as high-order perturbation source terms. Such high-order correction flows are used to depict: the density stratification effect caused by sharp variations of temperature and salinity; the local upwelling or downwelling flow caused by the variation of pycnocline thickness; the additional correction effect of thermal and saline stratification on the local flow direction and flow rate.

[0072] By sequentially superimposing the zero-order background flow, the first-order curvature correction flow and the high-order temperature and salinity gradient correction flow in the local area, a multi-scale approximate structure with clear physical layered meaning is formed, which provides local fine solution information for the construction of the overall scale equation.

[0073] Specifically, the formula of the multi-scale approximate structure is as follows:

[0074] wherein, represents the total velocity field of the ocean flow field; represents the zero-order dominant flow field; represents the first-order correction flow field; represents the second-order correction flow field; represents the Coriolis force parameter.

[0075] wherein, the zero-order dominant flow field The specific calculation formula is as follows: wherein, represents the time variation rate of the flow field; represents a horizontal gradient operator, which is used to depict the velocity gradient and variation of seawater in the horizontal plane; represents the tensor outer product of the gradient vector , which is used to construct the coupling term of the interface geometric convection field; a represents a scalar or tensor form of undetermined coefficient, which is used to depict the influence of interface curvature, pycnocline strength and other geometric terms on the flow field; g represents the equivalent function of the temperature and salinity pycnocline geometry; represents the zero-order pressure field; represents the zero-order external source term / driving term, which is the equivalent representation of background forcing, boundary driving or residual term; represents the velocity vector obtained by rotating counterclockwise by 90°.

[0076] S53: In the overall flow scale of the region, based on the multi-scale approximate structure, the multi-scale dynamics approximate expression is obtained with the Coriolis force parameter, the scale ratio parameter and the gradient characteristic as the dominant small parameters; the multi-scale dynamics approximate expression includes the effects of the earth rotation, the boundary geometry and the temperature and salinity stratification.​ In this embodiment, in the regional overall flow scale, the above-mentioned multi-scale approximation structure is taken as the encapsulated expression of local behavior, the earth rotation effect, the scale ratio parameter and the gradient characteristics and other small parameters are introduced, and the Navier-Stokes equation is expanded and simplified on the overall level.

[0077] 1. Introducing the Coriolis force parameter and the scale ratio parameter: In this process, the Coriolis force parameter related to the earth rotation is regarded as an important small parameter reflecting the effect of the earth rotation, which is used to describe the geostrophic balance and large-scale flow direction in the large ocean basin; at the same time, the characteristic length ratio and time scale ratio between the local scale and the overall scale are taken as the scale ratio parameter, which is introduced into the multi-scale expansion framework to distinguish the relative weight between the large-scale smooth flow field and the local rapid change flow field.

[0078] 2. Introducing the temperature and salinity gradient characteristics into the overall scale approximation expression: In addition to the geometric and rotation effects, the vertical and horizontal gradient characteristics of temperature and salinity are embedded in the overall flow equation in a parameterized form, which preserves the influence of the temperature and salinity stratification on the density, buoyancy and flow direction structure at the overall scale, so that the existence of the thermocline can still modulate the overall basin flow pattern in the large-scale expression.

[0079] 3. Obtaining an approximate expression containing multiple physical effects: Through the above processing, the multi-scale dynamic approximation expression obtained at the overall scale level contains: Earth rotation effect: embodied in the overall flow deflection and geostrophic balance structure; Boundary geometric effect: affecting the spatial distribution of the background flow through the curvature and fluctuation information of the equivalent geometric model; Temperature and salinity stratification effect: affecting the vertical shear and circulation structure of the flow velocity through the density stratification and buoyancy term.

[0080] The approximate expression becomes the basis for constructing the multi-scale dynamic approximation equation.

[0081] S54: Based on the multi-scale dynamic approximation expression, a multi-scale dynamic approximation equation is formed; the multi-scale dynamic approximation equation includes a Coriolis force term and a temperature and salinity coupling term.

[0082] In this embodiment, on the basis of the coupling of the local flow scale and the overall flow scale, the aforementioned multi-scale dynamic approximation expression is converted into a multi-scale dynamic approximation equation group that can be used for numerical solution and neural network constraint training, which contains: 1. Coriolis force term in momentum equation: used to describe the deflection effect of seawater under the background of the earth rotation, so that the model can reproduce the typical geostrophic flow structure and wind-driven flow field characteristics in the ocean.

[0083] 2. Terms coupled with temperature and salinity distribution: by introducing the spatial distribution of temperature field and salinity field into the momentum and mass conservation relationship, the temperature-salinity coupling term is formed, which is used to reflect the modulation effect of density stratification on the structure of the flow field, and the feedback effect of pycnocline intensity change on the flow velocity and direction.

[0084] In the equation construction, it is required that the modified flow of local scale is continuous in the pycnocline neighborhood and the background flow of global scale is continuous on the boundary, so that the zero-order background flow, the first-order curvature modified flow and the high-order temperature-salinity modified flow form a physically continuous and mathematically solvable integrated structure in the global flow field.

[0085] The obtained multi-scale dynamic approximation equation not only retains the basic physical connotation of Navier-Stokes equation, but also significantly reduces the difficulty of directly solving three-dimensional full complex equation through multi-scale decomposition, geometric equivalence and temperature-salinity stratification modeling, providing a feasible and stable physical model basis for subsequent constraint training (S6) based on physical information neural network.

[0086] S6: inputting the multi-scale dynamic equation and the temperature-salinity pycnocline equivalent geometric model into the physical information neural network for constraint training to generate a multi-scale flow field prediction model; Wherein, the step of inputting the multi-scale dynamic equation and the temperature-salinity pycnocline equivalent geometric model into the physical information neural network for constraint training to generate a multi-scale flow field prediction model comprises: S61: discretizing and coding the term structure of the multi-scale dynamic equation to input the physical information neural network in the form of physical residual; In this embodiment, the term structure of the multi-scale dynamic equation is discretized, including: Coriolis force effect term in the momentum equation; boundary geometric effect term affected by equivalent geometric curvature; multi-scale expansion term corresponding to zero-order background flow, first-order curvature modified flow and temperature-salinity high-order modified flow; stratification coupling term caused by temperature gradient and salinity gradient; mass conservation term and continuity constraint term of the global flow field.

[0087] In order to facilitate neural network training, the partial derivative structure of the above terms is converted into a physical residual form, that is, after discretizing the continuous dynamic equation at four-dimensional space points (longitude, latitude, depth, time), the physical deviation corresponding to each discrete point is taken as the input of the physical information neural network.

[0088] In specific implementation: a deep neural network with activation function is adopted as the physical information neural network subject; coordinates (x, y, z, t) of each sample point are taken as network input; network output contains multi-scale flow velocity components to be predicted and part of parameters (such as Ro, B0, B1); physical residual of each equation term is calculated by using automatic differentiation mechanism and taken as physical loss of the physical information neural network; the physical residual is minimized in training, so that the network prediction result automatically satisfies the physical structure of the multi-scale dynamic equation.

[0089] In this way, the dynamic equation is "coded into the network", so that the network learning process not only depends on data fitting, but also strictly follows the dynamic constraint.

[0090] Specifically, each term of the multi-scale dynamic equation is discretized to construct an approximate velocity expression of the physical information neural network, so as to calculate the residual; the formula of the approximate velocity expression of the physical information neural network is as follows: ; Among them, and represent the predicted value of the approximate horizontal velocity field output by the physical information neural network; and represent the reference flow; represents the first-order scale correction coefficient; represents the second-order scale correction coefficient; and represent the background flow velocity term.

[0091] Among them, the formula for calculating the physical residual of the physical information neural network is as follows: ; Among them, represents the reference flow velocity vector; represents the partial derivative of the reference flow with respect to time, i.e. the time evolution of the reference flow; represents the horizontal gradient operator; represents the Coriolis deflection effect; represents the pressure near the interface; h represents the background flow velocity near the interface.

[0092] S62: Based on the spatial structure characteristics of the temperature-salinity thermocline equivalent geometric model, the temperature field and salinity field observation data are processed in the same scale, and are taken as data constraint input into the physical information neural network; In this embodiment, the observed data (from the Copernicus Marine database) of temperature field, salinity field, flow rate field, etc. are resampled according to the depth distribution of the equivalent geometric model, so that the real data is consistent with the geometric model in the position of the thermocline and the curvature distribution; the data of different depth layers are mapped to the physical information neural network input space according to a unified scale (longitude-latitude-depth-time); the normalization processing is performed according to the variation amplitude of the data in the respective scale, so that the different physical quantities such as temperature, salinity and flow rate have comparability during training; the regularized data are input into the physical information neural network as data constraint items, including: temperature field and salinity field as boundary / supervision data; flow rate measured points (or reanalysis data) as velocity supervision constraint. This step ensures that the model not only complies with the dynamic structure, but also aligns with the actual thermocline geometric characteristics, and ensures that the predicted flow field structure has a real physical performance at the thermocline.

[0093] S63: Obtain a multi-scale flow field prediction model by iteratively optimizing the weight parameters of the physical information neural network.

[0094] In this embodiment, the weight parameters of the physical information neural network are gradually optimized through the superposition of multiple physical constraints and data constraints, and the loss structure includes the following types: Physical residual loss: make the flow rate and pressure field output by the network satisfy the structural relationship of the multi-scale dynamic equation.

[0095] Data fitting loss: make the predicted temperature, salinity and flow rate close to the real observed values.

[0096] Thermocline consistency loss: force the network to predict the flow field to be consistent with the curvature trend of the equivalent geometric model near the thermocline.

[0097] Direction loss: ensure that the deviation of the predicted flow direction from the real flow direction does not exceed the physical allowable range, and prevent the occurrence of "physically unreasonable flow direction mutation".

[0098] Momentum conservation loss and mass conservation loss: ensure that the prediction result meets the basic constraints of fluid dynamics in local and global scales.

[0099] Specifically, the network is trained by using an iterative optimization strategy, and the training steps include: 80% of the real data and physical points are used as the training set, and 20% are used as the validation set; In each round of training, the neural network weights are updated by back propagation; The physical residual is calculated in real time by automatic differentiation and fed back to the model; In the iterative process, the small parameters (such as Ro, B0 and B1) are automatically adjusted to meet the optimal multi-scale approximation structure; The training is terminated when the validation error and the physical residual both reach a set threshold.

[0100] The model obtained after the training has the following capabilities: according to the coordinates of any spatiotemporal point, the corresponding flow velocity and flow direction can be output in the entire sea area; in the vicinity of the thermocline, local acceleration zones and local deflection zones and other physical phenomena can be spontaneously reconstructed; the flow field structure presents continuity in the vicinity of the thermocline.

[0101] S7: Simulate and calculate the flow velocity and flow direction of the target sea area based on the multi-scale flow field prediction model, and compare with the measured flow field data to complete multi-scale modeling and verification.

[0102] In this embodiment, after the training of the multi-scale flow field prediction model is completed, the actual application and model verification link is entered. This step is divided into three parts: simulation calculation, data comparison, precision evaluation and model feedback.

[0103] (1) Simulation calculation: predict the three-dimensional flow velocity and flow direction field of the target sea area; Based on the physical information neural network (PINN) model, the boundary geometric information (i.e. the equivalent boundary structure of the thermocline and the initial temperature and salinity distribution field data) of the target time point are input to drive the model to output the three-dimensional flow velocity components (u, v, w) and the flow direction field of the entire target area at this time point.

[0104] The simulation output includes: Flow velocity modulus distribution map: express the velocity intensity change of each depth layer in the form of color gradient map.

[0105] Flow direction vector field map: draw an arrow map to represent the flow direction in the two-dimensional horizontal section (such as 100m, 300m, 500m depth).

[0106] Thermocline section velocity profile map: show the mutation of velocity above and below the thermocline along the vertical section, and capture the sensitivity of the fluid dynamic response to the thermocline structure.

[0107] The calculation adopts batch prediction mode to cover the entire target latitude and longitude and depth range, generating a complete spatiotemporal four-dimensional ocean flow field evolution sequence.

[0108] (2) Comparison and verification: compare the spatiotemporal error with the real observation data; To verify the precision and adaptability of the prediction model, multi-source real ocean observation data are selected as the control benchmark, including: Buoy profile data: obtain the buoy path data and the flow velocity observation values sampled at different depths in the sea area, and match the model output time node; Satellite flow field reanalysis data: obtain the sea current velocity distribution map of the surface layer (about 10m depth), and compare it with the predicted surface layer result horizontally; Mediterranean Ocean data assimilation products (such as CMEMS analysis data): used to compare the vertical velocity trend in the 100-600m depth interval.

[0109] Comparison method: align the model output and the observation value according to the same timestamp and spatial position; calculate the following error indicators: mean absolute error (MAE), mean absolute percentage error (MAPE), Pearson correlation coefficient (R) and root mean square error (RMSE).

[0110] The comparison shows that the model performs significantly better in the thermocline dense area (200-400m), and the predicted flow velocity trend is highly consistent with the observed profile data, with a maximum relative error of less than 7.3%; in the surface area, the correlation coefficient with the satellite flow field reanalysis data reaches 0.93.

[0111] (3) Model feedback: error diagnosis and prediction ability analysis; For the local error hot spot area (such as the area with steep submarine topography or high-frequency internal wave disturbance) that appears in the comparison process, the error source of the model is analyzed: check whether the sample density of the corresponding area in the training data set is sparse; analyze the error propagation effect of the Gaussian curvature boundary; detect whether the partial derivative gradient of the residual term in the PINN in this area is saturated or oscillating.

[0112] Further, an integrated multi-model average method is used to integrate multiple independently trained models with different initial weights to improve the prediction robustness in uncertain areas.

[0113] Finally, the spatiotemporal stability of the model in the entire prediction period is verified, and the following key conclusions are summarized: the model can accurately predict local flow rate mutations caused by the thermocline and boundary shear flow; the velocity error within 100m above / below the thermocline is significantly smaller than that of traditional flow field interpolation models; the model output has the engineering characteristics of visualization, interpretability and applicability, and has the potential to be deployed as a marine forecasting module.

[0114] Reference Figure 3 As shown in the figure, the figure contains four subgraphs, which respectively show the comparison relationship between the horizontal flow velocity components u and v predicted by the physical information neural network constructed based on the application and the real marine reanalysis data and the residual distribution thereof, for verifying the prediction accuracy and physical consistency of the multiscale flow field prediction model of the application under the action of the thermocline.

[0115] The left upper graph is a scatter plot of the horizontal flow velocity east component u, the horizontal axis is the real data, the vertical axis is the model prediction data, and the red dashed line is the reference line of the predicted value completely consistent with the real value in the ideal state. From the graph, it can be seen that the scatter points are highly concentrated near the reference line, indicating that the predicted value and the real value have good consistency; in the low flow rate area ( The predicted points (ranging from 0.05 m / s to 0.05 m / s) are continuously distributed without significant systematic shifts, indicating that the model can accurately characterize subtle velocity changes near the mesophase. In locally high-velocity regions, the model maintains a stable linear relationship without significant extrapolation bias. This demonstrates that the physical information neural network prediction based on multi-scale dynamic equations and equivalent geometric models can effectively learn eastward background flow, curvature-corrected flow, and higher-order perturbation structures driven by temperature-salinity gradients.

[0116] The upper right figure is a scatter plot comparing the northward component of the horizontal flow velocity v. It can be observed that most predicted points are distributed along the reference line, showing a clear positive correlation trend; however, when the actual values ​​are relatively small ( Within the mainstream flow range (0.05 to 0.03 m / s), the prediction results more closely match the reference line, indicating that the model performs well even in weak flow fields. A few deviations typically occur in areas of dramatic local undulations or abrupt gradient changes, related to the local curvature of the equivalent geometric model, but the overall deviation is small. This demonstrates that the present invention can effectively handle local flow field disturbances caused by changes in the geometry of the tiered flow path and maintain the stability and continuity of flow direction prediction under multi-scale approximation structures.

[0117] The lower left figure shows the residual distribution (u_pred) between the predicted and actual values ​​of the u component. The features in the figure include: the residuals exhibit an approximately symmetrical bell-shaped distribution, indicating that the prediction error does not have a directional shift; the residuals are mainly concentrated in... In the range of 0.02 to 0.02 m / s, the peak value is obvious and the high-value region is narrow, indicating low and stable error. The residual distribution decays smoothly at the tail, without long tails or skewness, indicating that the model will not suffer from serious deviations due to complex boundaries (such as high-curvature transition regions). This further verifies that the physical information neural network effectively suppresses overfitting and maintains dynamic consistency during training through multiple physical constraints such as momentum conservation, mass conservation, and orientation loss.

[0118] The bottom right figure shows the prediction residuals of the v component (v_pred). The statistical distribution of v_true is shown in the figure. It can be observed that the residuals also exhibit an approximately Gaussian distribution, with the center clearly concentrated near zero; the main residual intervals are concentrated in... The accuracy of the prediction for the v component is similar to that of the u component, with a range of 0.015 to 0.015 m / s. The high peak in the distribution indicates that the error at many prediction points is extremely small, demonstrating that the model of this invention has a highly consistent predictive capability in both longitudinal and latitudinal directions. The symmetry and high concentration of the residual distribution indicate that this model can reasonably respond to flow deflection near the thermohaline, density stratification effects, and local vortex disturbances, conforming to the actual dynamic structure in the physical scenario.

[0119] Combining the above four figures can be concluded that the multiscale flow field prediction model constructed by the application shows high prediction ability on both u and v components; the good alignment of scatter plots and the concentrated residual distribution fully prove the physical consistency and numerical stability of the model.

[0120] Please refer to Figure 4 The embodiment of the application also provides a multiscale modeling and verification system based on oceanic jump layer structure analysis, which comprises: A data acquisition module is configured to acquire parameter data of a temperature-salinity jump layer region in a target sea area. A preprocessing module is configured to perform outlier rejection and spatial interpolation on the parameter data to form a temperature-salinity jump layer data set. A feature extraction module is configured to extract temperature-salinity jump layer spatial structure features based on the temperature-salinity jump layer data set by using a convolutional neural network. A first construction module is configured to quantize temperature-salinity jump layer boundary complexity by using a Gaussian curvature parameter based on the temperature-salinity jump layer spatial structure features, and establish a temperature-salinity jump layer equivalent geometric model. A multiscale decomposition module is configured to perform multiscale decomposition on a Navier-Stokes equation set by taking the temperature-salinity jump layer equivalent geometric model as a boundary condition, and generate a multiscale dynamic approximation equation. A second construction module is configured to input the multiscale dynamic equation and the temperature-salinity jump layer equivalent geometric model into a physical information neural network for constrained training, and generate a multiscale flow field prediction model. A verification module is configured to simulate and calculate flow velocity and flow direction of the target sea area based on the multiscale flow field prediction model, and compare with measured flow field data to complete multiscale modeling and verification.

[0121] It can be understood that the contents in the above method embodiments are all applicable to the present system embodiment, the present system embodiment specifically realizes the same functions as the above method embodiments, and achieves the same beneficial effects as the above method embodiments.

[0122] The embodiment of the application also provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor realizes the above method when executing the computer program. The electronic device can be any intelligent terminal, such as a tablet computer or a vehicle-mounted computer.

[0123] It can be understood that the contents in the above method embodiments are all applicable to the present device embodiment, the present device embodiment specifically realizes the same functions as the above method embodiments, and achieves the same beneficial effects as the above method embodiments.

[0124] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method.

[0125] It can be understood that the contents in the method embodiments are applicable to the storage medium embodiments, the storage medium embodiments specifically implement the functions of the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0126] The embodiment of the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the method.

[0127] It can be understood that the contents in the method embodiments are applicable to the program product embodiments, the program product embodiments specifically implement the functions of the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0128] The memory is a non-transitory computer readable storage medium, which can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0129] The embodiment of the present application provides a multi-scale modeling and verification method and system based on ocean current jump layer structure analysis. The scheme obtains parameter data of the temperature and salinity jump layer region of a target sea area, performs outlier rejection and spatial interpolation, effectively improves the data quality, and provides a reliable foundation for subsequent modeling. Secondly, the spatial structure features of the temperature and salinity jump layer are automatically extracted by using a convolutional neural network, avoiding the limitations of artificial feature design, and enhancing the expression ability of complex jump layer forms. Further, the jump layer boundary complexity is quantified by using a Gaussian curvature parameter, and an equivalent geometric model is constructed to provide accurate boundary conditions for multi-scale dynamic modeling. In addition, the Navier-Stokes equation set is decomposed in multiple scales, and the jump layer structure is combined to constrain the input physical information neural network for training, realizing a multi-scale flow field prediction model with strong physical consistency and high calculation efficiency. Finally, the prediction result is compared and verified with the measured data to ensure the accuracy and practicability of the model. The scheme realizes the automation, refinement and multi-scale coupling of the jump layer structure modeling, and improves the scientificity and engineering applicability of the ocean current modeling.

[0130] The embodiments described in the specification are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0131] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0132] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0133] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the functions of the modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0134] The terms "first", "second", "third", "fourth" and the like (if any) in the specification of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0135] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but this does not limit the scope of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and spirit of the present application should be within the scope of the present application.

Claims

1. A multi-scale modeling and verification method based on ocean current strata structure analysis, characterized in that, The method includes: Obtain parameter data for the thermo-salinity gradient region of the target sea area; The parameter data is subjected to outlier removal and spatial interpolation to form a thermo-salinity gradient dataset; Based on the aforementioned temperature-salinity transition layer dataset, spatial structure features of the temperature-salinity transition layer are extracted using a convolutional neural network. Based on the spatial structural characteristics of the thermo-salinity transition layer, the boundary complexity of the thermo-salinity transition layer is quantified by the Gaussian curvature parameter, and an equivalent geometric model of the thermo-salinity transition layer is established. Using the equivalent geometric model of the thermo-salinity stratum as boundary conditions, the Navier–Stokes equations are decomposed into multi-scale equations to generate multi-scale dynamic approximation equations. The multi-scale dynamic equations and the equivalent geometric model of the thermo-salinity transition layer are input into the physical information neural network for constraint training to generate a multi-scale flow field prediction model. Based on the multi-scale flow field prediction model, the flow velocity and direction of the target sea area are simulated and calculated, and compared with the measured flow field data to complete the multi-scale modeling and verification.

2. The method according to claim 1, characterized in that, The parameter data includes temperature distribution, salinity gradient, ocean current velocity, and depth information.

3. The method according to claim 1, characterized in that, The step of extracting spatial structure features of the temperature-salinity transition layer using a convolutional neural network based on the temperature-salinity transition layer dataset includes: The temperature-salinity gradient dataset is preprocessed into a grid, and the temperature field and salinity field are constructed into a multi-channel input tensor according to the depth hierarchy. A convolutional neural network is used to perform multi-scale convolution extraction on the multi-channel input tensor to obtain local spatial features characterizing the gradient changes of the thermo-salinity transition layer. Based on the local spatial features, a feature mapping sequence is generated by pooling and feature compression operations on the convolution output; The feature mapping sequence is input into a fully connected structure to generate spatial structure features.

4. The method according to claim 1, characterized in that, The spatial structural features include the location of the thermo-salinity transition layer boundary, gradient intensity distribution, and thickness variation information of the transition layer.

5. The method according to claim 1, characterized in that, The step involves quantifying the boundary complexity of the thermohaline using Gaussian curvature parameters based on its spatial structural characteristics, and establishing an equivalent geometric model of the thermohaline, including: The spatial structural features are discretized and reconstructed to obtain the initial surface of the layer boundary; Based on the initial surface, the Gaussian curvature of each discrete point is calculated, and curvature anomalies are smoothed to generate the curvature distribution of the jump boundary. The complexity of the mezzanine boundary is quantified based on the curvature distribution of the mezzanine boundary. The quantification results are used as equivalent geometric constraints to construct an equivalent geometric model of the thermo-salinity transition layer; the equivalent geometric model is used to characterize the equivalent curvature of the transition layer boundary and its spatial variation trend.

6. The method according to claim 5, characterized in that, The formula for the Gaussian curvature is as follows: ； Where G represents the Gaussian curvature of the thermohaline interface; This represents the first-order spatial rate of change of the thermohaline interface along the longitude direction; This represents the first-order spatial rate of change of the thermohaline interface in the latitudinal direction. This represents the second-order curvature of the thermohaline interface along the longitude direction; This represents the second-order curvature of the thermohaline interface along the latitudinal direction; This represents the mixed curvature term of the thermohaline interface in the longitude-latitude direction.

7. The method according to claim 1, characterized in that, The process of using the equivalent geometric model of the thermohaline as boundary conditions to perform multi-scale decomposition of the Navier–Stokes equations generates multi-scale dynamic approximation equations, including: Based on the curvature distribution of the equivalent geometric model of the thermo-salinity layer, the Navier–Stokes equations are divided into local flow scales and regional global flow scales in the neighborhood of the thermo-salinity layer. In the local flow scale, the interface curvature characteristics, temperature-salinity gradient changes and the thickness parameter of the stratum are introduced into the mechanical equation as boundary disturbance source terms to construct a multi-scale approximation structure; the multi-scale approximation structure includes a zero-order background flow, a first-order corrected flow dominated by interface curvature and a higher-order corrected flow driven by temperature-salinity gradient. Within the overall flow scale of the region, based on the multi-scale approximation structure, a multi-scale dynamic approximation expression is obtained with the Coriolis force parameter, scale ratio parameter, and gradient characteristics as the dominant small parameters; the multi-scale dynamic approximation expression includes the Earth's rotation effect, boundary geometry effect, and temperature-salinity stratification effect. Based on the aforementioned multi-scale dynamic approximation expression, a multi-scale dynamic approximation equation is formed; the multi-scale dynamic approximation equation includes a Coriolis force term and a temperature-salinity coupling term.

8. The method according to claim 7, characterized in that, The formula for the multi-scale approximation structure is as follows: in, This represents the total velocity field of the ocean current field; This represents the zeroth-order dominant flow field; This represents the first-order corrected flow field; This represents a second-order corrected flow field; This represents the Coriolis force parameter.

9. The method according to claim 1, characterized in that, The step of inputting the multi-scale dynamic equations and the equivalent geometric model of the thermo-salinity gradient into a physical information neural network for constraint training to generate a multi-scale flow field prediction model includes: The term structure of the multi-scale dynamic equation is discretized and encoded, and then input into the physical information neural network in the form of physical residuals. Based on the spatial structure characteristics of the equivalent geometric model of the temperature-salinity gradient, the temperature field and salinity field observation data are normalized at the same scale and used as data constraints input into the physical information neural network. By iteratively optimizing the weight parameters of the physical information neural network, a multi-scale flow field prediction model is obtained.

10. A multi-scale modeling and verification system based on ocean current strata structure analysis, characterized in that, The system is configured to perform the method as described in any one of claims 1-9, the system comprising: The data acquisition module is used to acquire parameter data of the thermo-salinity stratum region in the target sea area; The preprocessing module is used to remove outliers and perform spatial interpolation on the parameter data to form a thermo-salinity gradient dataset. The feature extraction module is used to extract the spatial structure features of the thermo-salinity transition layer based on the thermo-salinity transition layer dataset using a convolutional neural network. The first construction module is used to quantify the boundary complexity of the thermo-salinity transition layer by using Gaussian curvature parameters based on the spatial structural characteristics of the thermo-salinity transition layer, and to establish an equivalent geometric model of the thermo-salinity transition layer. The multi-scale decomposition module is used to perform multi-scale decomposition of the Navier–Stokes equations using the equivalent geometric model of the thermo-salinity layer as boundary conditions, and to generate multi-scale dynamic approximate equations. The second construction module is used to input the multi-scale dynamic equation and the equivalent geometric model of the thermo-salinity stratum into the physical information neural network for constraint training, and generate a multi-scale flow field prediction model. The verification module is used to simulate and calculate the flow velocity and direction of the target sea area based on the multi-scale flow field prediction model, and compare it with the measured flow field data to complete the multi-scale modeling and verification.

Images